I'm Head of AI at Myria
Biosciences, where AI-designed microbes produce macrocyclic-peptide therapeutics. I lead the
machine-learning and data platforms that turn the company's discovery and design data into
decisions.
I build the models and platforms that make living cells predictable — and, ultimately, programmable.
More broadly, my work spans the two halves of biological modelling that rarely meet:
mechanistic, mathematical models of how cells work — whole-cell simulation, dynamical systems,
resource allocation — and machine learning that predicts biological function directly from sequence
and data. I believe the next advance in AI-for-biology comes from fusing the two, wrapped in autonomous
agents that design and run their own experiments — toward an AI-driven virtual cell you can query,
steer, and trust.
Previously I earned my PhD at the University of Edinburgh (deep learning for protein expression), with
earlier degrees from Imperial College London and Harvard.
A self-driving ML system that automatically builds, critiques, and improves models predicting expression or yield from short biological sequences — for any such dataset, not one hand-tuned model. An LLM council proposes approaches, reviewers challenge them, and a bounded ML-engineer loop validates each under fixed metrics. Already spans five public datasets (~1.12M sequences across E. coli, yeast, and human).
A platform that opens a molecular-resolution whole-cell E. coli model (~1,592 genes) to a natural-language agent: design knockout and media-shift experiments, launch batch simulations, and analyse results by conversation, with human approval for any state change — putting whole-cell modelling in the hands of experimental biologists and students, not just computational specialists.
The Well, for the Cell In development
A public ML benchmark of whole-cell dynamics — the biology counterpart to Polymathic AI's The Well. A large, standardized corpus of whole-cell E. coli simulation trajectories (fully observable, causally perturbed, dynamic) with benchmark tasks, leakage-free splits, and baselines: model-derived data that complements the sparse, noisy snapshots of experimental omics. The wcEcoli platform generates the corpus; the Coli agent is its grounded query interface.
A grounded, guardrailed Claude copilot over a whole-cell E. coli simulation, built for the Built with Claude: Life Sciences hackathon. It answers only from real simulation data and enforces biosecurity and scientific guardrails — refusing experiments outside the model's validated envelope and withholding non-viable results rather than laundering them into clean-looking numbers.
